Abstract
This note points out that the model order selection criteria proposed earlier for Davison's and Marshall's reduced order models are not applicable in many of the practical cases. A new criterion is proposed to overcome the deficiencies of the earlier methods.